I'm hitting a regression with LTP's pwritev02 [1] on s390x with 4.8.0-rc7.
Specifically the EFAULT case, which is passing an iovec with invalid base 
  #define CHUNK 64
  static struct iovec wr_iovec3[] = {
        {(char *)-1, CHUNK},
hangs with 100% cpu usage and not very helpful stack trace:
  # cat /proc/28544/stack
  [<0000000000001000>] 0x1000
  [<ffffffffffffffff>] 0xffffffffffffffff

The problem starts with d4690f1e1cda "fix iov_iter_fault_in_readable()".

Before this commit fault_in_pages_readable() called __get_user() on start
address which presumably failed with -EFAULT immediately.

With this commit applied fault_in_multipages_readable() appears to return 0
for the case when "start" is invalid but "end" overflows. Then according to
my traces we keep spinning inside while loop in iomap_write_actor().

Patch below makes the issue go away for me:

diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index 66a1260b33de..e443dbd2b5df 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -600,11 +600,11 @@ static inline int fault_in_multipages_readable(const char 
__user *uaddr,
                                               int size)
        volatile char c;
-       int ret = 0;
+       int ret = -EFAULT;
        const char __user *end = uaddr + size - 1;

        if (unlikely(size == 0))
-               return ret;
+               return 0;

        while (uaddr <= end) {
                ret = __get_user(c, uaddr);



Reply via email to